Published on: 8 September 2023 | Last updated: 29 November 2023
Inclusion criteria
Sexual violence captured within the ACLED dataset includes all political or public violence of a sexual nature. This means that sexual violence included in the dataset is not limited to solely penetrative rape, but also includes actions like public stripping, sexual torture of men, etc.1 “The following acts, among others, are considered sexual violence: (a) forcing a person to undress in public; (b) sexual penetration; (c) rape; (d) sexual molestation.” See International Committee of the Red Cross, ‘Practice relating to Rule 93. Rape and Other forms of Sexual Violence,’ last accessed 24 February 2023; “An act of a sexual nature is not limited to physical violence, and may even not involve any physical contact — forced nudity is an example of the latter.” See International Criminal Court, ‘Policy Paper on Sexual and Gender-Based Crimes,’ June 2014 Especially during ‘war time,’ but also in periods of political instability more broadly, the use of sexual violence as a strategy to reinforce power structures is not uncommon. As such, what is known as ‘sexual violence in conflict’ or ‘conflict-related sexual violence,’ such as war-time rape, are also included in the dataset. Additionally, the ACLED dataset also includes other forms of sexual violence targeting any individual – regardless of gender or age – perpetrated by armed, organized actors.
The ACLED dataset does not include sexual violence stemming from domestic, interpersonal, or intimate partner violence occurring outside of the political/public sphere; these events are outside of ACLED’s mandate, and such violence (sexual or not) is not captured within the ACLED dataset.
How sexual violence is recorded in the ACLED dataset
ACLED is an event-based dataset, meaning each entry in the dataset is an ‘event.’ Events are denoted by the involvement of designated actors, occurring in a specific named location and on a specific day.2 For further information regarding ACLED’s unit of observation, see the ACLED Codebook. Additionally, ACLED event types and sub-event types are hierarchical to accommodate for the recording of concurrent tactics within the same event, in order to avoid double-counting. This means that a ‘Sexual violence’ event (e.g. a civilian is raped) that occurred within the same context as a ‘Mob violence’ event would be coded as one ‘Mob violence’ event. As such, the ‘Sexual violence’ sub-event type3 The ‘Sexual violence’ sub-event type was introduced to the ACLED dataset in March 2019, when ACLED released its new sub-event type categorization. exclusively captures sexual violence events under the ‘Violence against civilians’ event type, whereas sexual violence in the context of other forms of violence coded higher in the ACLED event type hierarchy – like Riots – is marked with the “sexual violence” tag in the ‘Tags’ column.4 For more on this, see the Definitions of ACLED Event and Sub-Event Types section in the ACLED Codebook.
Further considerations
When recording sexual violence, an event can involve one to many victims. One person sexually assaulted by a soldier in a specific town on a certain day would be coded as a single event. An episode of mass rape by an armed militia reported in a specific town on a certain day would be recorded in the same way. The number of ‘Sexual violence’ events should, therefore, not be conflated with the number of sexual violence victims – in the same way that the number of violent events in the ACLED dataset should not be conflated with the number of fatalities.
Collecting accurate data on violent events is difficult due to a lack of detailed, verified reporting during active violence. Further, the count of victims of violent events – whether counting fatalities or casualties, or the number of sexual violence victims specifically – is often the most biased and poorly reported component of data around political violence. These numbers can vary widely, especially as there can be incentive to overstate or underreport these numbers by the parties engaged in the violence.
Counting sexual violence by number of victims alone results in making areas where this type of violence is more readily reported – as a result of structural or social restrictions – appear more susceptible to sexual violence, and areas with fewer victims reporting such violence appear less dangerous. Coding such violence by event rather than by the number of victims is, therefore, a step towards minimizing this implication. Individuals and entities responsible for reporting also have intended and unintended biases to inflate numbers to misrepresent the size of groups, in order to illicit support and move international bodies to action, or minimize international backlash, etc. When the number of victims is mentioned in reporting, this information is recorded in the ‘Notes’ section of the event within the ACLED dataset; users can choose to use this information if they wish while understanding the caveats above.
Specifically in the context of sexual violence, underreporting by victims is common due to backlash or normative concerns. Events involving a single victim who does not report the sexual violence they experience would not appear in the dataset at all. Given the nature of such events, events with a single victim are more likely to go underreported than events with a large number of victims. Coverage within the ACLED dataset, as in all datasets, is limited to what has been reported. ACLED tries to capture an accurate picture of political violence through the use of various sources of reporting: traditional media, 5 ACLED researchers review thousands of traditional media sources in over 20 languages ranging from national newspapers to local radio. new media, 6 New media refers specifically to sources such as trusted Twitter accounts (e.g. journalists) and vetted Telegram channels.reports by international organizations, or information gathered by local partners. However, sexual violence specifically, perhaps even more so than other forms of violence, can suffer from underreporting as a result of, though not limited to, fear of repercussions, legal restrictions, and psychological trauma. These limitations should be considered when drawing conclusions from the data.
